The Non-local And Block Matching 3-D Filtering Algorithm Research

De-noising of noisy images is always an important issue in image processing, and there has been increasing attention to it since the higher image quality requirement needed in imaging analysis and follow-up applications. However, most de-noising algorithms proposed before show outstanding performance only to some particular image models but fail in general, and they always remove image fine structures or create artifacts. The nonlocal-means (NLM) algorithm developed by Buades et al. has successfully solved these problems, and furthermore, the Block Matching and 3-D Filtering (BM3D) algorithm, which is also based on the nonlocal idea and exploits filtering in the spectrum domain, reaches much better results both in subjective evaluation and objective quality standards. The BM3D algorithm is regarded as the current state-of-the-art performing de-noising algorithm.To both of the NLM algorithm and the BM3D algorithm, there is a big problem of the computational burden with the image-block-matching, which has already affected their applications in practice. Therefore, speeding up these algorithms is of paramount importance to us now. There have been many methods proposed to improve the efficiency of the NLM algorithm, and the work in this paper is to find some methods from them which are also effective on the BM3D algorithm.We firstly review the evolution of de-noising methods by using a new classification. Then we propose an optimized version of the BM3D algorithm, including the semi-local BM3D algorithm and the mean/variance-BM3D algorithm.We present some experiments and simulations comparing the optimized method and the original one, which all verify the efficiency of this algorithm developed in this paper.In the end of the paper, the direction of our future research is given.